Prediction of Indian summer monsoon rainfall using Niño indices: A neural network approach

It is an established fact that sea surface temperature (SST) anomalies in the central-eastern Pacific associated with the El Niño-Southern Oscillation (ENSO) act as predominant forcing of the All India Rainfall Index variability. However, the same has been found to be difficult to simulate. In the present study, we have attempted to improve the seasonal forecast skill of the Indian Summer Monsoon Rainfall Index (ISMRI). Correlation analysis is done to see the effect of SST indices of Niño-1 + 2, Niño-3, Niño-3.4 and Niño-4 regions on ISMRI with a lag period of 1–8 seasons. Significant positive correlations, with confidence level above 99%, are found between ISMRI and (i) Niño-3 index with a lag of 4 (June–July–August) and 5 (March–April–May) seasons, (ii) Niño-3.4 index with a lag of 4 and 5 seasons and (iii) Niño-4 index, with a lag of 5 seasons before the onset of monsoon. These SST indices are used for prediction of ISMRI using multiple linear regression and Artificial Neural Networks (ANNs) models. A comparative examination of the results suggests that the ANN model has better predictive skills than all the linear regression models investigated, implying that the relationship between the Niño indices and the ISMRI is essentially non-linear in nature.

Research highlights
► Correlation analyses of Indian summer monsoon rainfall with Niño indices is done.
► It is found that ISMR has significant lead-lag relation with Niño indices.
► Five predictors (SST indices) are selected for the prediction of ISMRI.
► Multiple linear regression and Artificial Neural Network (ANN) models are employed.
► ANN model predicts significantly better implying that the above relation is non-linear.